A match is not an infringement
The most useful mistake we've made so far: a domain our scanner scored as a near-certain clone, that turned out to be a legitimate business on another continent.
The near miss
The domain carried the brand's exact word and sold physical goods. Scored high. A human looked before anything was filed — and found a flooring company in another country, sharing the brand's word in its own language, selling floors. An automated pipeline would have sent that notice, and a wrongful notice is not a rounding error: it exposes the client, and us, to real legal risk.
We tried to teach the machine. It didn't take.
We wrote two rules to catch the case automatically. One let the innocent company through but also let through a real clone that hadn't started selling yet. The other caught the clone and kept flagging the innocent company. The honest conclusion: this is not a decision a machine should make. The scanner now marks the doubt — “this may be a different line of business” — and a person decides.
The same lesson, on Amazon
A search for one brand's signature product returned a full page of results without the brand itself on it. A padding vendor would report every one of them as an infringement. Most were legitimate businesses in the same category. Counting them makes a report longer, not truer.
What this buys you
Three kinds of look-alikes we have caught in testing and let go: a similar silhouette, a competitor in your category, and a business that happens to share your word in another language. Every finding that does reach you has cleared the three gates — right, evidence, authority — and is one you can act on.
What a first scan actually finds
61 live domains behind one apparel brand. Ten were shops. Sixteen were parked. One was a flooring company.
How long a copy survives, by channel
11.4 days on TikTok Shop against 3.2 on Amazon in the published data — and why we still won't promise you a removal date.